The Contribution of Stakeholders in
Different Phases of the Fraud Process
Simone Goijaerts University of Groningen
Abstract. This study examines the contribution of stakeholders into different phases of the fraud process. Several studies reveal that financial statement fraud evolves over time due to the phenomenon of the slippery slope. Different stakeholders can be involved in fraudulent reporting such as the CEO, CFO, controllers and VPs. Conflicting interests of these stakeholders might impact the development of the fraud process and influence the (number of) stakeholders involved in fraudulent financial reporting. Multiple studies offer useful theoretical insights concerning the way in which fraud can be normalized in an organization over time using the fraud phases. However, they did not test this model empirically and did not discuss the stakeholders involved in these phases of fraud in their studies. This study therefore addresses this gap by studying 186 U.S. fraud cases. Hand-collected data is used to test the hypotheses. The data is gathered from the SEC litigation database for the period 2000 - 2015. The results reveal that the CEO is the initiator while being involved. Non C-level employees can be the initiator of fraudulent financial reporting as well. The manipulation techniques used in financial statement fraud differ dependent on the process stage and are different when other stakeholders become involved. The number of stakeholders involved increases when the duration of the fraud increases. New stakeholders furthermore contribute to fraudulent reporting due to enlargement and/or replacement. These results of this study can be valuable for the prevention and detection of fraudulent financial reporting for managers, audit committees, external auditors and regulators in the future.
Keywords: Fraudulent Financial Reporting, Stakeholders, Fraud Process, US Complaints.
2
The Contribution of Stakeholders in
Different Phases of the Fraud Process
Combined thesis Accountancy and Controlling
Simone Goijaerts
University of Groningen
Faculty of Business and Economics January 21, 2019
Zernike Campus, Duisenberg building Groningen, Netherlands, 8747 AE (06)230 755 37 E-mail: s.c.goijaerts@student.rug.nl Student number: S2364069 Simone Goijaerts Postjeskade 85-3 1058DJ Amsterdam Phone number: 06-81368718 Supervisor: dr. K. Linke Co-assessor: dr. J.S. Gusc
3 TABLE OF CONTENT
INTRODUCTION ... 4
THEORETICAL BACKGROUND AND HYPOTHESES DEVELOPMENT ... 7
Stakeholder Theory. ... 7 Process Model ... 8 Hypothesis Development ... 10 Initiation Phase ... 10 Proliferation Phase ... 12 Institutionalization Phase ... 13 Socialization Phase ... 14 Discovery Phase ... 14 METHODOLOGY ... 17 Type of Research ... 17 Data Source ... 17 Data Collection... 18 Sample ... 18 Variables ... 19 Control Variables ... 22
Big Four Firm ... 22
Financial Crisis ... 22
Firm Size ... 22
Data Analysis and Assumptions ... 23
RESULTS ... 24 Descriptive Statistics ... 24 Testing Hypotheses ... 25 Additional Analysis ... 34 DISCUSSION ... 39 CONCLUSION ... 42 REFERENCES ... 44 APPENDIX I ... 50 APPENDIX II... 55 APPENDIX III ... 56 APPENDIX IV ... 57
4
INTRODUCTION
“All organizations are inherently criminogenic” (Gross, 1978, p 56). White collar crime is inherent to all types of industries, all levels within an organization and both in small and large companies (McGee and Byington, 2017). White collar crime is defined as the performance of illegal acts by organizational professionals, which involve amongst other things cheating, lying and fraudulent reporting (Sutherland, 1939). Over the last decade, white collar crime became a crucial social and economic problem that had influence on many different stakeholder groups. Financial statement fraud is increasingly carried out by employees within an organization (Ashfort and Anand, 2003). More specifically, financial statement fraud committed by C-functions (e.g. CEO, CFO, COO) has become a considerable concern for organizations (Darvall-Stevens, 2015). A CEO is able to influence the decision-making process and thereby, the performance of the organization (Adams, Almeida and Ferreira, 2005). Beasley, Carcello, Hermanso and Neal (2010) state that not only the C-level managers (e.g. CEO, COO and CFO) commit financial statement fraud, other individuals such as auditors, controllers, VPs1 and
lower level employees are involved as well. For instance, the CFO and the controller were the initiators of the fraud at WorldCom (Cunningham, 2003). Former top executives (among other CFO and the vice chairman) cooked the books at Cendant Corporations2.
Besides the contribution of stakeholders in financial statement fraud, Schrand and Zechman (2012) argue that unethical behavior resulting in this fraud evolves over time. Indeed, Beasley et al. (2010, p. 3) find that “most fraud cases were not isolated to a single fiscal period”. They find that the average fraud has a duration of 31 months. In addition, Beasley, Carcello and Hermanson (1999) show that most fraud cases include at least two fiscal periods. This suggests that once a company commits fraud, the company carries on with committing the fraud for multiple periods.
Welsh, Ordonez, Snyder and Christian (2015) argue that financial statement fraud evolves over time due to phenomenon what they call ‘the slippery slope of unethical behavior’. This means that people commit small unethical acts, which over time may lead to people committing larger unethical acts. Mostly, it starts with an optimistically biased misstatement which is not
1 This is an overall name for VPs of different departments such as the VP sales, VP finance, VP Operations and so
on.
5 intentional (Schrand and Zechman, 2012). This misstatement can be identified as optimistic ex post which can lead to an increasing probability of misstatement due to overconfident executives. Many major scandals such as Enron started with small indiscretions and grew over time (Kirchner, 2010; Welsh et al., 2015). Ashfort and Anand (2003) state that financial statement fraud can be seen as a slippery slope where initial corrupt practices become normalized and institutionalized in the organization.
Current empirical studies are however incomplete concerning the contribution of different stakeholders (i.e. CEO or CFO) in financial statement fraud evolving over time. Beasley et al. (1999) find that the fraud has a duration of two years. It starts with small manipulations and grows over time. Besides, Gino and Bazerman (2009) argue that people are more likely to accept unethical behavior of their colleagues when this kind of behavior develops over time instead of all of a sudden. In addition, Schrand and Zechman (2012) show that financial statement fraud is executed for three years without describing the way in which the fraud develops in these three years. All of these studies suggest that financial statement fraud evolves over time but fail to describe the phases in which the fraud develops. In contrast, Ashfort and Anand (2003) and Palmer and Maher (2006) focus in their studies on the (cognitive) processes to engage in fraudulent financial reporting. They state that fraud develops over time in different phases. While both studies offer useful insights concerning the way in which fraud can be normalized in an organization over time using these phases, the authors did not discuss the organizational players involved in these phases of fraud. This study addresses this gap by studying 186 U.S. fraud cases. This area of research has up to now not been fully explored, the specific contribution of this study will be explained below. The main research question of this study is the following: ‘What is the contribution of stakeholders in different phases of the fraud process in an organization?’
Hand-collected data is used to test the hypotheses. The data is gathered from the SEC litigation database for a sample of 186 U.S. firms for the period 2000 - 2015. The results of this study show that the CEO is the initiator of the fraud and at the same time involved in the fraud perpetration. The manipulation techniques used in financial statement fraud differ dependent on the process stage and are different when other stakeholders become involved. The number of stakeholders involved increases when the duration of the fraud increases. New stakeholders furthermore contribute to fraudulent reporting due to enlargement and/or replacement.
6 This study contributes to existing literature on fraudulent financial reporting by combining the fraud phases of the studies of Ashfort and Anand (2003) and Palmer and Maher (2006). In addition, this study adds an extra fraud phase to this combined process model. It is furthermore the first time that this model is empirically tested. Next to that, this research adds to the existing literature by studying the contribution of several organizational players on each phase of the fraud process. These findings may help with the prevention and the detection of financial statement fraud by external auditors and audit committees.
The remainder of this paper is structured as follows. The next section will describe the theoretical background and the development of the hypotheses. After that, the methodology and the sample selection will be explained. Then the results of this study will be outlined. Finally, the results will be discussed, and the paper will be concluded.
7
THEORETICAL BACKGROUND AND HYPOTHESES DEVELOPMENT Stakeholder Theory. The stakeholder theory is a widely used theory to define the stakeholder
concept and to classify stakeholders into different categories (Rowley, 1997). A stakeholder is defined as “any group or individual who can affect or is affected by the achievement of the organization's objectives” (Freeman, 1984, p. 46). This definition is one of the broadest definitions of a stakeholder in the existing literature (Mitchell, Agle and Wood, 1997). Freeman and Reed (1983, p. 91) outline a narrower definition of stakeholders by defining stakeholders as groups “on which the organization is dependent for its continued survival”. An organization has both external and internal stakeholders (Darnall, Seol and Sarkis, 2009). External stakeholders can be the auditor, the society, suppliers or customers. Internal stakeholders consist of management and non-management employees (Waddock and Graves, 1997).
The stakeholder theory can be used to define various objectives of an organization (Ansoff, 1965). One important objective is to balance inconsistent demands of multiple internal and external stakeholders of an organization (Roberts, 1992). Following this theory, organizations should pay attention to the needs and interests of all of their stakeholders rather than only e.g. their shareholders (Jawahar and McLaughlin, 2001). All stakeholders, both internal and external, have different and therefore often conflicting interests (Reed, Graves, Dandy, Posthumus, Hubacek, Morris, Prelle, Quinn, Stringer, 2009). Employees prefer high wages, pensions and other (financial) benefits while customers want high quality products for low prices. Shareholders want high dividend payments while investors want low risks and high returns on investments (Jensen, 2002). This illustrates that a CEO may have different personal interests, has to account for many stakeholder interests compared to for example a sales manager. Specific interests of one stakeholder might benefit him or herself but harm another stakeholder (Sternberg, 1997). Organizations must consider the needs and interests of stakeholders to maximize firm performance and to achieve the strategic objectives of an organization (Roberts, 1992; Länsiluoto et al., 2013).
Conflicting interests between an organization and its stakeholders might impact the development of the fraud process (Demski, 2003) and influence the (number of) people involved in fraudulent financial reporting. These conflicting interests might negatively affect the continuity of the organization. Therefore, it is important for an organization to understand and to balance these conflicting interests to maximize firm performance and to reduce the fraud risk (e.g. Demski, 2003).
8
Process Model. There are different theories or models describing the process of corporate
corruption. These theories can be applied to financial statement fraud which is a form of corporate corruption. Financial statement fraud develops through different phases or processes. The theory of a process model can be based on the work of Brief, Bertram and Dukerich (2001), Ashfort and Anand (2003) and Palmer and Maher (2006). Brief et al. (2000) and Ashfort and Anand (2003) analyze collective corruption. Both studies argue that fraudulent financial reporting is executed by multiple organizational members. They explain the way in which collective corruption becomes institutionalized and normalized in an organization. Ashfort and Anand (2003) describe this phenomenon of normalization by using three pillars. These pillars are 1) institutionalization, 2) rationalization and 3) socialization. Institutionalization is the process by which an initial corrupt practice turns into a routine. Rationalization is the process by which an individual justifies his/her self-serving behavior and legitimizes his/her actions. Socialization is the process by which new people in an organization learn to perform the corrupt practices (Ashfort and Anand, 2003).
Brief et al. (2001) describe the process of ongoing corporate corruption by highlighting several circumstances such as initial collective compliance, institutionalization, socialization and the way in which collective wrongdoing can be curbed. Brief et al. (2001, p. 477) define compliance as “the initial obedience of a collective of employees to an official authorization to engage in a corrupt practice”. Corrupt practices can be institutionalized by the “collective interpretation” of acts which are “ethically loaded”. This interpretation justifies these wrongful corrupt practices. The authors provide many policy prescriptions for hindering wrongdoing by organizational members. Most of these result from moral instruction and corporate governance reform (Palmer, 2008).
Ashfort and Anand (2003) and Brief et al. (2001) implicitly provided a process model which consists of four phases (Palmer and Maher, 2006; Palmer, 2008). These four phases are the initiation phase, proliferation phase, institutionalization phase and socialization phase. The initiation phase is the phase in which top managers decide to commit a corrupt practice. In the proliferation phase, other employees become involved in the corrupt practice besides the initiators(s) of fraudulent reporting. These corrupt practices become embedded in organizational routines and norms in the institutionalization phase. In the socialization phase, newcomers in an organization learn techniques to support corrupt behavior and practices. Palmer and Maher (2006) describe five recommendations in which this process model can be
9 extended. These recommendations relate to a rational cost-benefit analysis, social influence processes, a combination of these two and contextual factors (Palmer and Maher, 2006). One of these recommendations refers to the initiator of the fraud. Lower level managers or employees might be the initiator of corrupt practices as well rather than only top managers. This recommendation is used in this thesis.
Brief et al. (2001), Ashfort and Anand (2003) and Palmer and Maher (2006) indicate that collective corruption evolves over time through different phases. Essential to their process models is something what researchers call the slippery slope of unethical behavior (i.e. Welsh et al., 2015). This slippery slope means that multiple small violations can lead to larger unethical moves in preparing financial statements. People are more likely to accept unethical behavior of their colleagues when unethical behavior develops over time instead of all of a sudden (Gino and Bazerman, 2009). Indeed, individuals in an organization have the tendency to justify small unethical indiscretions and are prone to carry out small unethical acts which are mostly not intentional (Welsh et al., 2015). These individuals find it easier to rationalize small unethical practices rather than abrupt unethical practices which have a huge impact (e.g. Mazar, Amir and Ariely, 2008). Participants3 in the study of Suh, Sweeney, Linke and Wall (2018) illustrates this slippery slope by finding it difficult to remember and to define the very first moment they became involved in (small) fraudulent financial reporting. Therefore, this study expects that there exists a slippery slope of unethical behavior by organizational participants meaning that there are no conscious decisions made that lead directly to fraudulent financial reporting.
3 All participants were high level executives with a C-function who were involved in enormous financial statement
10
Hypothesis Development. Financial statement fraud can result in serious consequences for
both the organization and its stakeholders (Bhasin, 2013). Most frauds are committed by people within the organization who are in a position of trust, such as the C-functions or the accounting or finance department (Carpenter et al., 2011). Indeed, PwC (2018) reports that the main perpetrators of fraud are internal stakeholders (52 percent). 40 percent4 of the frauds are mostly committed by external stakeholders. Examples of these external stakeholders are customers, vendors or agents. These are often third parties who are ‘frenemies’ of the firm and with whom the firm has a business relationship (PwC, 2018). This study will focus on the stakeholder(s) who contribute to the fraud process. Interesting questions are which stakeholder(s) is/are the initiator(s) of fraudulent financial reporting? Do more stakeholders contribute to the fraud process over time? If so, which stakeholders? Who is the detector of financial reporting fraud?
The process model used in this study consists of five phases. This model includes the first recommendation of Palmer and Maher (2006) and adds an extra phase to this model which is the discovery phase. These five phases are 1) initiation phase, 2) proliferation phase, 3) institutionalization phase, 4) socialization phase and 5) discovery phase. These phases are explained below.
Initiation Phase. The initiation phase involves around the initiator of and the manipulation technique used in financial statement fraud. This phase concerns the person or persons who decided to engage in fraudulent financial reporting. Dunkelberg and Jessup (2001) state that the initiator of misreporting obtained a successful position and is in a position of authority. Due to a high level of authority, the initiator was not suspected by other people in the organization. One of the persons that has likely attained a position of trust and authority in a firm is the CEO. His or her power is undoubtedly taken for granted if their power is institutionalized in the organization (Shen, 2003). A CEO is then able to put pressure on lower level managers and employees to commit corrupt practices (Clinard, 1983). The CEO has the power to influence decisions and thereby the financial performance of a company (Adams et al., 2005). Beasley et al. (1999) state that in the most cases the CEO was the main initiator of fraudulent financial reporting. The CEO is named by the Accounting and Auditing Enforcement Releases (AAERs) in 72 percent of the fraud cases. Brennan and McGrath (2007) finds similar results and shows that the CEO is the main initiator in 12 out of 14 studied fraud cases. Beasley et al. (2010) find the same results compared to their study in 1999: the CEO is the initiator in 72% of the fraud
11 cases. Next to that, the involvement of the CFO increased significantly over the years. The AAERs associated the CFO with fraudulent financial reporting in 65 percent of the fraud cases in their study of 2010 compared to 43 percent in their study in 1999.
In contrast to the findings mentioned above, Palmer and Maher (2006) describe that corrupt practices can also be initiated by lower level managers and employees. Beasley et al. (2010) argue furthermore that not only the CEO and CFO are important perpetuators of fraud, financial middle managers (FMM), controllers and VPs are named by the AAERs as well. The FMM are named in 34 percent of the alleged cases while the VPs are associated with fraudulent reporting in 38 percent of the cases. A survey held by the Controllers Institute showed that 31 percent of the controllers sometimes manipulate accounting numbers5. Even 24 percent of the controllers
admit that manipulating accounting numbers happens more often. These FMM operate at an operational level and are accountable for accounting and reporting tasks. These tasks and responsibilities make them able to commit illegal acts in the financial statements and budgets. Therefore, this study expects that the CEO will be the main initiator based on the studies of Brennan and McGrath (2007) and Beasley et al. (1999; 2010). Based on the study of Palmer and Maher (2006), lower level managers or employees might initiate fraudulent financial reporting as well. Therefore, these hypotheses 1a and 1b are formulated in the following way:
Hypothesis 1a: In the fraud process, the CEO is the main initiator of financial statement fraud given the CEO is involved.
Hypothesis 1b: In the fraud process, lower level managers or employees are the main initiators of financial statement fraud.
The second part of this phase concerns the manipulation method the initiator is using. The following categories are identified by the SEC6: improper revenue recognition, improper expense recognition and improper or omitted disclosure, lying to the auditor and improper asset valuation. According to the studies of Beasley et al. (1999, 2000), Bonner et al. (1998) and Linke (2012) improper revenue recognition7 is the most commonly used technique. This is
5https://www.accountant.nl/globalassets/accountant.nl/blad/2010-nr1. 12 / acc_2010_12_ creatief_met
_controllers.pdf
6 SEC publication in the SOX act of 2002.
7 Improper revenue recognition can be categorized in subcategories: improper timing of revenue, fictitious revenue,
12 followed by improper expense recognition and improper disclosure of information to mislead others. Linke (2012) shows in her study that the manipulation of revenues and expenses occurs more frequently over the years8. It therefore is useful to analyze the manipulation method(s) used in the financial statements. Having more information about the manipulation method increases the knowledge of which manipulation method is used and the main drivers of stakeholders to manipulate the revenues, expenses or assets (Linke, 2012). Due to different interests and job responsibilities of these stakeholders, the manipulation method may differ. For example, CEOs in contrast to CFOs may manipulate revenues in the income statement to demonstrate sales growth (Feng, Ge, Luo and Shevlin, 2011). Their compensation package may be determined based on this sales growth and thus reported earnings. CFOs are mostly responsible for financial reporting, budgeting and the financial planning (Kaufman, 2003; Gore et al., 2007). If a budget is overstated or understated, the CFO is able to inflate expenses or earnings by providing accounting schemes to meet the budget (Feng et al., 2011). Therefore, the following hypotheses are formulated:
Hypothesis 2: The manipulation method used in financial statement fraud may differ dependent on the process stage.
Hypotheses 3: The manipulation method used in financial statement fraud may differ when other stakeholders become involved.
Proliferation Phase. This phase enlists the support of others in the fraud process. Many large-scale fraud scandals in the last decade are carried out by more than one person (Free and Murphy, 2015). According to Ramamoorti (2008, p. 529), fraudulent financial reporting becomes a “team sport and it often includes collusion”. Indeed, “financial fraud is a team sport” (FD, 07-10-20189). For example, the chairman of the board, CFO and CEO were blamed for cooking the books at Enron Corporation (Rezaee, 2015). The financial department i.e. CFO and controller and other financial managers were accused for financial statement fraud at WorldCom. And at Aurora Foods Inc., the CEO, CFO, customer financial services manager and senior financial analysts were involved in fraudulent financial reporting. Therefore, researchers have recently focused more on the co-offenders of fraud. According to Rezaee (2005), financial statement fraud is committed by a group of knowledgeable and intelligent
8 When compared with the studies of Bonner et al. (1998) and Beasley et al. (1999, 2000). 9https://fd.nl/ondernemen/1272868/financiele-fraude-is-een-teamsport
13 perpetuators such as high-level executives or auditors. The combination of the CEO, CFO and another person are involved in most of these recent fraud scandals (Free and Murphy, 2015). Indeed, the AAERs state that in 89 percent of the cases, the CEO and the CFO together are the main initiators of misreporting (Beasley et al., 2010). This number is increased with 6 percent compared to the 83 percent of the fraud cases in 1999. It is not surprising that more people are involved in fraud due to the fact that one single person does not have all the skills, resources and access to engage in fraudulent financial reporting (COSO 2010; Free and Murphy, 2015). Therefore, this study expects that more people will become involved in fraudulent financial reporting over time. Therefore, the following hypothesis is formulated:
Hypothesis 4: The number of stakeholders involved in fraudulent reporting will grow in the proliferation phase.
Institutionalization Phase. In this phase, the committed fraud is embedded in organizational routines and cultural norms. The organization’s culture might influence an organization to consider financial statement fraud (Geriesh, 2003). The culture affects the attitude of an organization towards fraudulent financial reporting (Watson, 2003), i.e. organizations can have a culture of ‘aggressive reporting behavior’. Culture can be adopted as a routine in the organization as corrupt practices are repeated and become habitual (Ashfort and Anand, 2003). Routinizing can be defined as “transforming the action into routine, mechanical, highly programmed operations” (Kelman, 1973, p. 46). If fraudulent behavior is embedded and routinized in cultural norms, it strongly affects various subunits and levels in an organization. Ashfort and Anad (2003, p. 4) state that fraudulent financial reporting “can become an integral part of the day-to-day activities to such an extent that individuals may be unable to see the inappropriateness of their behaviors”. A corrupt act which is part of a (daily) routine can be split into more specific tasks executed by different individuals. These individuals may perform their tasks without knowing how their individual actions, in conjunction with the actions of others, contribute to the enactment of a corrupt practice (Bunderson, 2001). Individuals are not critical anymore when incorrect information is provided (Linke, 2012). The following hypothesis is developed:
Hypothesis 5: Fraudulent financial reporting is institutionalized in routines and cultural norms of stakeholders in an organization.
14 Socialization Phase. This is the phase in which newcomers in an organization learn techniques to support corrupt behavior and practices. In this study, a newcomer is defined in two different ways. Firstly, a newcomer can be defined as a person that replaces someone engaged in the fraud process. Secondly, a newcomer can be defined as someone who becomes involved with an enlargement of the fraud scheme. Many organizations use ‘onboarding’ to facilitate the socialization process and to help socialize new employees (Klein, Polin and Sutton, 2015). Onboarding is defined as “the formal and informal practices, programs, and policies enacted or engaged in by an organization or its agents to facilitate newcomer adjustment” (Klein and Polin, 2012, p. 268). In the onboarding process, newcomers will be introduced to their jobs, will get to know the organization culture and will become familiar with organizational norms, value, goals and rules (Watkins, 2016). The quicker new employees in an organization know the organizational culture and firm specific knowledge, the quicker newcomers can contribute to organizational success and create a competitive advantage for their firm (Coff and Kryscynski, 2011). However, this onboarding process often provides newcomers with much information which is extremely difficult to incorporate in a short period of time for new employees (Bradtand and Vonnegut, 2009). Therefore, top management and other high-level executives are important in this phase. High level executives set the ethical tone of behavior and this will be pushed down through the organization. Newcomers in a firm search for guidelines to behave in an acceptable manner. These high-level executives are role models for providing these guidelines for ethical behavior (Dunkelberg and Jessup, 2001). Due to the fact that new employees perceive them as role models, newcomers become easily involved in fraudulent financial reporting if these high-level executives are involved. Therefore, the following hypothesis is expected:
Hypothesis 6: New stakeholders contribute to fraudulent financial reporting in the socialization phase.
Discovery Phase. This phase indicates the detector of fraudulent financial reporting. The detector of fraud is the person that discovered the fraud in an organization for the first time and brought this to light. The fraud can be detected by people inside the organization such as executives, non c-level employees, a whistle-blower or the internal auditor. Besides, the fraud can be detected by an external auditor, the media or an analyst who are all external parties. Research of Dyck, Marse and Zingales (2010) show that the key detectors of fraud are employees of all levels in a firm, analysts and the media, respectively in 18.3, 16.9 and 15.5
15 percent of the fraud cases. An auditor detects 10.5 percent of the fraud cases according to Dyck et al. (2010). The aim of external auditors is to provide reasonable assurance that the financial statements of firms are free from material misstatements (Jizi, Nehme and ELHout, 2017). External auditors can detect fraudulent financial reporting when they test the financial statements although they are not responsible for the detection and reporting of this fraud (e.g. Haniffa and Hudaib, 2007; Alleyne and Howard, 2005). The internal audit is an important tool in the detection of fraudulent financial reporting by insiders in the organization (Nestor, 2004). Internal auditors have superior firm specific knowledge and are able to spend much time on the detection of fraud (Coram, Ferguson and Moroney, 2008). C-level and non C-level employees such as financial middle managers, controllers or VPs can be trained to recognize indicators of potential fraudulent reporting to reduce fraud risks (Samociuk, Doody, and Iyer, 2010). They must be aware of ‘red flag’ behavior of people in the organization. Higher debts, long absences from work and changes in work patterns are examples of red flag behavior. Next to that, the media can identify fraudulent financial reporting due to the fact that journalists collect and analyze information for their readers of e.g. news blogs or newspapers (Dyck et al., 2010). A journalist might benefit from the detection of fraud if it results in a better reputation. Analysts can detect fraud due to the fact that they are trained in analyzing and interpreting information about firms (Dyck et al., 2010). Moreover, a whistleblower can discover fraudulent financial reporting. Eaton and Akers (2007, p. 67) define whistleblowing as: “the act of reporting wrongdoing within an organization to internal or external parties”. Eaton and Akers (2007) argue that a whistleblower can either report information concerning fraudulent financial reporting to a source inside the company (internal whistleblowing) or report information to a source outside the company (external whistleblowing). For these reasons, the parties identified in this phase are: the internal and external auditor, management of an organization, the media, an analyst, a whistleblower or other (third) parties. However, most frauds are detected by internal stakeholders within the specific organization in an early stage (Dyck et al., 2010; Kaplan, Pope, and Samuels, 2011). Internal stakeholders such as employees might be aware of fraud earlier than others external auditors or analysts due to tips from other employees within the organization. Other employees might obtain knowledge of the fraudulent financial reporting or might have been asked or forced by executives to engage in fraudulent financial reporting (ACFE, 2010). The involvement of multiple stakeholders might therefore influence internal detection. Due to the fact that more (internal) stakeholders are involved, more people in the organization are aware of the committed fraudulent practices. Leading to a higher chance of
16 these illegal acts being reported by an internal stakeholder. Therefore, the following hypothesis is formulated:
Hypothesis 7: In the detection phase, fraudulent financial reporting is detected internally when more stakeholders contribute to the fraud process.
Conceptual model Fraud process Initiation phase CEO involvement Fraud technique used The number of stakeholders Proliferation phase Institutio-nalization phase Socialization phase Detection phase Cultural norms New stakeholders Internal detection
17
METHODOLOGY
Type of Research. In order to be able to research the contribution of stakeholders in the
different phases of the fraud process this study utilizes a quantitative research approach. This study will use archival data to be able to answer the main research question. Archival data is often used for researching financial statement fraud (i.e. Beasley et al., 1999; 2010). Archival data therefore seems useful for answering the research question of this study. Moreover, Linke (2012, p. 13) argues that “the availability of data and prior research models guides the choice of a quantitative research approach”. A large amount of literature regarding financial statement fraud already exists. The theories outlined in the existing literature will act as a theoretical foundation for the research model of this study. For example, the normalization of corruption in firms (Ashfort and Anand, 2003), the (cognitive) processes to commit financial statement fraud (Palmer and Maher, 2006) and the analysis of financial statement fraud occurrences affecting U.S. public companies (Beasley et al., 1999; 2010).
Qualitative research such as case studies or interviews might be able to answer the research question of this study as well. Linke (2012) however argues that these qualitative methods have some practical risks. She mentions that it is very probable that individuals who have been involved financial statement fraud, either committing or witnessing it, are reluctant to participate in these kinds of interviews. Adding to this is the fact that the majority of financial statement fraud cases occurred and are well-documented in the United States (ACFE, 2012). There is thus a large risk on non-responses and convincing people who are not eager to respond, to eventually participate in an interview is unpractical and costly for a researcher located in the Netherlands (Linke, 2012). Combining these difficulties will probably lead to a small number of interviews which make (statistically) answering the research question a hard thing to do. Furthermore, the use of archival data allows for the empirical testing of the theoretical model put forward by Ashfort and Anand (2003 and Palmer and Maher (2006) which has up to now not been validated empirically. This study therefore chooses to use a quantitative research method; the use of archival data.
Data Source. Most information is gathered from Accounting and Auditing Enforcement
Releases (AAERs) issued by the Securities and Exchange Commission (SEC). These litigation releases are commonly used in financial statement fraud research (i.e. Dunn, 2004; Beasley et al., 1999; 2010). The SEC complaints are used for the reason that the nature of the fraud is well-documented in detail (Bonner et al., 1998). In addition, the documentation i.e. the complaints
18 is publicly available. The use of these complaints and litigation releases involves several disadvantages. For example, there might exist a potential bias regarding the enforcement strategy of the SEC (Linke, 2012). Besides, it is difficult to prove guilt since many parties settle the allegations (Dunn, 2004; Beasley et al., 2010). Despite these disadvantages, these complaints and litigations are the best available data source to study the research question. Compustat is used as an additional database, data with regard to the total assets of fraud companies is collected from this database.
Data Collection. My supervisor handed me a dataset of companies involved in fraud, me and
4 fellow students then found the AAERs matching these fraud cases. From these AAERs the data is then hand-collected by the 5 of us. We divided the total number of cases by five and studied the cases separately. We agreed on a deadline for completing the dataset and created one excel file containing the complete dataset. When someone was in doubt regarding one of the complaints, that specific complaint was discussed by the five researchers. However, due to 5 different researchers collecting data, there might be some inconsistency in the dataset. Data is gathered from listed companies in the US, since information on fraud cases in the United States is publicly available (Linke, 2012). Next to that, the list of fraud cases which occurred in the US is quite extensive (Bhasin, 2013). According to the report of the ACFE (2012), 57.2 percent of the fraud cases were companies originated in the United States, this amounts to a total of 778 fraud cases. In countries other than the US, it is not possible to collect this amount of quantitative data. The SEC has the enforcement authority to bring enforcement actions against organizations and individuals for violating the securities laws10.
Sample. The sample of this study includes cases of fraudulent financial reporting which are
reported in the SEC database. The initial sample comprises 233 fraud cases in the period January 1, 2000 to December 31, 2015. January 1, 2000 is chosen as starting point since it takes time until financial statement fraud is actually detected. There are furthermore limited complaints and litigation releases available earlier than year 2000. December 31, 2015 is chosen as end point since it takes time to study all these complaints and litigation releases. Moreover, the age ofdigitization started around the year 2000. Prior to the year 2000, systems of firms were less automated and therefore the implementation and control of internal controls were more difficult.
I only selected cases of intentional financial statement fraud. These cases are selected
19 conforming the definition of Beasley et al. (1999, p. 11). They define financial statement fraud as: “the intentional material misstatement of the financial statements or financial disclosures or the perpetration of an illegal act that has a material direct effect on the financial statements or financial disclosures”. Next to that, this study only includes individuals who are alleged to have executed fraudulent financial reporting. Firms who are alleged rather than individuals are excluded: 47 cases were excluded. The final sample contains 186 fraud cases. However, there exists a difference in the sample size per phase due to the duration of the fraud process. The sample size is lower in the proliferation and institutionalization phase since some fraud processes have no proliferation or institutionalization phase. I therefore excluded respectively 14 and 35 fraud cases for the proliferation and institutionalization phase resulting in a final sample size of 172 and 151 cases. In addition, some sample sizes are lower due to missing values. An example is the sample size for the detector of fraud. The variables relating to the detector of fraudulent reporting have a sample size of 128 cases since many cases did not actually report the detector of fraud. A reason for this might be that some parties do not admit their wrongdoing but instead settle the allegations. By using this sample, I expect to find sufficient and well-documented cases of financial statement fraud cases.
Variables. The variables in this study relate to the different phases in the fraud process based
on studies of Brief et al. (2001), Ashfort and Anand (2003) and Palmer and Maher (2006). The process model of fraudulent reporting takes place over time through these different phases of fraud and is established in terms of quarter years. The initiation stage starts at period t = 0, which is the period in which the fraud actually starts. The proliferation phase includes the second and third quarters of the fraud. The institutionalization and the socialization phase cover the quarters four till the quarter prior to the detection of the fraud. The institutionalization and socialization phases might be intertwined which makes it extremely difficult to distinguish these two phases in the complaints. Therefore, these phases are combined. Lastly, the detection phase includes the quarter in which the fraud is discovered. All phases are dummy variables. The phases have a value of one if the fraud takes place in that specific quarter and a value of zero if not. The first variable used in this study to test hypothesis 1a is the involvement of CEO (CEO_INV) which is used to test whether the CEO is involved in the fraud process or not. This is scored as a dummy variable due to the fact that it is difficult to test the strength of involvement of the CEO. This variable has a value of one if the CEO is involved in the fraud process and a value of zero otherwise. Another variable used to test hypotheses 1 is the CEO as initiator of fraudulent financial reporting (CEO_INI). This is scored as a dummy variable. CEO has a value
20 of one if the CEO is the initiator fraud process and a value of zero otherwise. The CEO is determined as initiator when he or she is involved in the initiation phase or put pressure on lower level managers and employees to commit financial statement fraud. The variables used in this study to test hypothesis 1b is the involvement of non C-level employees (EMP_INV) and non C-level employees as initiator (EMP_INI). Both variables are scored as dummy variables due to the fact that it is again difficult to test the strength of involvement. Employees are determined as initiators when they are involved in the initiation phase of the fraud.
To test hypotheses 2, the fraud technique used to commit fraudulent reporting are split in five categories: improper revenue recognition (TECH_REV), improper expense recognition (TECH_EXP) and improper or omitted disclosure (TECH_DIS), lying to the auditor (TECH_LY) and improper asset valuation (TECH_ASS). All three variables are measured as dummy variables. For example, REV has a value of one if improper revenue recognition is used as manipulation technique.
Hypothesis 3 tests whether the manipulation method may differ when other stakeholders become involved by using the variables DIFF_TECH and DIFF_STAKE. Variables are included for the number of new manipulation techniques used and the number of new stakeholders involved in each stage (e.g. DIFF_TECH_PRO and DIFF_STAKE_PRO). These are numerical variables. When there are no new techniques used or stakeholders involved, it is scored as a zero. If there is one extra manipulation technique used or stakeholder involved, it is scored as a one. If there are two extra manipulation techniques used or stakeholders involved, it is scored as a two and so on.
To test hypothesis 4, the number of people involved in the initiation phase (NUMB_INI) and the number of people involved in the proliferation phase (NUMB_PRO) are used. These variables are numerical.
To test hypothesis 5, cultural norms (CULT) is a dummy variable and tests whether the group of people committing fraud remains stable or not. Stable is defined as the number of stakeholders involved remains the same. It has a value of one if this group remains stable and it has a value of zero if not. The number of stakeholders involved in each phase of the fraud process (i.e. NUMB_INI, NUMB_PRO, NUMB_INS, NUMB_DET) are used to test whether the group of people committing the fraud is stable or not to test hypothesis 5. These variables are numerical.
To test hypothesis 6, newcomers can be defined in two ways: (1) a newcomer is a person that replaces someone engaged in the fraud process (NEW_REP) and (2) a newcomer is someone
21 who becomes involved with an enlargement of the fraud scheme (NEW_ENL). Both are dummy variables.
Hypothesis 7 includes the variables regarding to the detector of fraudulent financial reporting. This study distinguishes nine different detectors including the variables: the internal auditor (DET_INTAUD), the external auditor (DET_EXTAUD), management (DET_MAN), VPs (DET_VPS), the media (DET_MED), analysts (DET_ANA), whistleblower (DET_WHI), the audit committee (DET_AUDCOM) or other (third) parties (DET_OTH). Again, these variables are dummy variables. To test this hypothesis, the variables number of stakeholders involved (NUMB_INV) and internal detection (DET_INT) are used. As said earlier, NUMB_INV is a numerical variable. DET_INT is a dummy variable and is scored as 1 if the detector is an internal stakeholder and zero otherwise.
22
Control Variables. Prior studies show that some control variables relate to the dependent
variable. The control variables used in this study are related to auditor characteristic (the type of audit firm), the influence of the financial crisis and the size of the fraud firm. This section explains these control variables.
Big Four Firm. Auditor characteristic is explained by the type of firm. Two types of firms can be distinguished: a big four audit firm and a non-big four audit firm. Prior studies show that firms audited by a big four firm have a lower incidence of financial statement fraud (e.g. Lennox and Pittman, 2010). Fraudulent financial reporting is more likely to be detected by big four audit firms rather than non-big audit firms due to the fact that big four firms provide a higher audit quality than non-big four firms (DeAngelo, 1981; Rezaee, 2005). In addition, big four firms provide better monitoring and assurance services to their own clients compared to non-big four firms (Farber, 2005; Francis, 2004). Next to that, these firms have more expertise in auditing large public firms, have more industry-specific knowledge and have more resources such as advanced technology and high-skilled personnel (Carpenter and Strawser, 1971; Rezaee, 2005). Lastly, big four audit firms are more likely to withstand pressure from large clients (Rezaee, 2005). Auditor type (BIG4) is measured using a dummy variable coded 1 if the auditor is from a big four audit firm. It is coded 0 otherwise. A negative relation is expected between fraudulent financial reporting and a big four firm.
Financial Crisis. The recent financial crisis affected fraudulent reporting by threatening the capital markets (Black, 2010). According to Fligstein and Roehrkassea (2016, p. 617): “the financial crisis of 2007 to 2009 was marked by widespread fraud”. Mishkin (2011) divides the financial crisis in two phases: the subprime mortgage crisis in 2007 and the global financial crisis in 2008 and 2009. The global financial crisis intensified 2008 due to many collapses such as Lehman Brothers. These collapses resulted in a loss of confidence (Kirkpatrick, 2009). During the recent global financial crisis, firms were not able to balance the different interests of stakeholders (Karaibrahimolu, 2010). The result was that firms behaved conservatively and mainly opportunistically. Opportunistic behavior of firms in the financial statements often results in fraudulent reporting (Fligstein and Roehrkassea, 2016). Therefore, fraudulent financial reporting might occur more during the financial crisis. The financial crisis (FINCRI) is a dummy variable. It has a value of 1 when the committed fraud is executed in the years 2008 and/or 2009 and a 0 otherwise.
Firm Size. The last control variable concerns the size of fraud firms. Fraudulent financial reporting occurs in firms of all sizes (Beasley et al., 2010). However, larger firms relative to
23 small firms are better able to implement cost-effective internal controls. These internal controls might reduce the likelihood of or prevent fraudulent financial reporting (Beasley et al., 1999). Larger firms might voluntarily improve the structure of the audit committee by adding more outside directors or executives with knowledge related to financial statement fraud (Beasley and Salteria, 2001). Smaller firms are unwilling or might not have knowledge and resources to implement such internal controls or to improve their audit committee. To overcome the problem of heteroscedasticity and to reduce the likelihood of outliers, firm size is measured as the natural logarithm of total assets (i.e Chakrabarty, 2001; Beasley et al., 1999). The total assets are in millions of US dollars.
Data Analysis and Assumptions. All dummy variables will be tested by using a chi-square
test. When testing a numerical and a dummy variable, a logistic regression will be used. The results of the regressions will be explained in the next section. A linear regression has four key assumptions 1) a linear relationship, 2) normality, 3) homoscedasticity, and 4) little or no multicollinearity (Field, 2017). A chi-squared has two assumptions: 1) independence and 2) sample size. Independence means that each subject or observation is included in only one cell. A chi-squared only works well if the sample size is large enough (Field, 2017). If the sample size for each cell is lower than five, the Fisher exact test will be a more appropriate measure to use. Violation of these assumptions result in a reduction of power. Other assumptions apply to logistic regressions: 1) the observations must be independent of each other meaning that there are no repeated measures, 2) little or no multicollinearity, 3) the sample size must be greater than ten observations (Field, 2017). A paired-sample t-test has two assumptions, namely 1) observations must be independent (i.e. paired observations must be drawn randomly) and the scores between two variables must be normally distributed (Field, 2017). These assumptions will be checked and reported in the results section.
24
RESULTS
Descriptive Statistics. Table 1 shows the descriptive statistics for this study, which contains
fraud cases from the period 2000 – 2015. The average fraud has a duration of 7.87 quarters. There are multiple techniques used to commit fraud which are, including their occurrence, the following: 1) improper revenue recognition (67%), improper expense recognition (40%), 3) improper or omitted disclosure (61%), 4) lying to the auditor (55%) and improper asset valuation (16%). There exists a change in the manipulation method used in 69% of the fraud cases. Next to that, the number of stakeholders contributing to fraudulent reporting is increased since in 53% of the fraud cases new stakeholders contribute to the fraud process. Most new stakeholders become involved in the proliferation phase of the fraud process. There are almost 3 stakeholders involved on average. New stakeholders might become involved due to replacement (15%) or enlargement (17%). In 61% of the fraud cases, the auditor is working at a big four audit firm. The fraud took place during the financial crisis in 19% of the fraud cases.
Table 1: Descriptive Statistics
N Minimum Maximum Mean Std. Deviation
DUR 186 1 25 7.866 5.860 TECH_INI 186 0 5 1.467 0.721 TECH_PRO 172 0 5 1.814 0.937 TECH_INS 151 0 5 2.172 1.204 TECH_DET 186 0 5 2.000 1.286 TECH_REV 186 0 1 0.667 0.473 TECH_EXP 186 0 1 0.403 0.492 TECH_DIS 186 0 1 0.608 0.490 TECH_LY 186 0 1 0.548 0.499 TECH_ASS 186 0 1 0.156 0.364 DIFF_TECH 186 0 1 0.688 0.464 DIFF_STAKE 186 0 2 0.527 0.511 NUMB_INV 186 1 10 2.882 1.754 CULT_NUMB 151 0 1 0.570 0.497 NEW_ENL 151 0 1 0.166 0.373 NEW_REP 151 0 1 0.146 0.354 NEW_SOC 151 0 1 0.238 0.428 BIG4_AUD 180 0 1 0.611 0.489 SIZE 150 0.846 18.801 7.542 3.876 FINCRI 186 0 1 0.188 0.392 Valid N (listwise) 118
25
Testing Hypotheses. Hypothesis 1a. The involvement of the CEO and the CEO as initiator
have a significantly high correlation (r = 0.974) meaning that these variables vary together (Table 2, Panel A). The crosstabulation is demonstrated in Panel B to gather more information. Theoretically, there exists no independence problem since the CEO can be involved and can be the initiator of the fraud at the same time. The CEO might also be the initiator of the fraud by putting pressure on other employees and therefore not being involved directly. However, in almost all cases the CEO is involved and is the initiator of the fraud as well. In zero cases the CEO is the initiator of the fraud while not being involved and in only five cases the CEO is involved but is not the initiator of fraudulent financial reporting. From the crosstabulation it can be seen, that if the CEO is the initiator, he or she is always involved. I therefore can conclude that there is still an independence problem in this hypothesis meaning that the assumption of independence is not met.
The other assumption relates to the sample size. This assumption is met since zero cells have an expected count less than five: the minimum expected count is 40.19 as shown in Panel C. Panel C shows the results of the chi-square test and shows that there is a significant association between the involvement of the CEO and the CEO as initiator X2(1) = 166.94, p < 0.001. The predictability of this model is high, since in 97%11 of the cases the model correctly predicts whether the CEO is involved or not which is the same for CEO being the initiator. The CEO is in 84 out of 186 cases not involved in the fraud process. However, if the CEO is involved than he or she is the initiator in 97 cases out of 102 cases. Based on these results, I can conclude that the CEO is only the initiator when he or she is involved in the fraud process. Therefore, this hypothesis can be accepted with the limitation that independence is restrained.
11 (84 + 97) / 186 = 97.3%
26 Hypothesis 1b. The involvement of employees and employees acting as an initiator have a significant high correlation (r = 0.689) meaning that these variables vary together. A crosstabulation then provides more information and is shown in table 3, Panel A. The assumption of independence is not met: a non C-level employee is only initiator when he or she is involved as well. The other assumption relates to the sample size. This assumption is met since zero cells have an expected count less than five: the minimum expected count is 30.79 as shown in Panel C. This Panel shows the results of the chi-square test, which reveals that there is a significant association between the involvement of non C-level employees and non C-level employees being an initiator X2(1) = 99.39, p < 0.001. The predictability of this model is high, since the model in 82%12 of the cases correctly predicts whether the employees are involved or not which is the same for an employee being the initiator. In 69 out of 186 cases the employee
12 (69 + 83) / 186 = 81.7%
Table 2 Panel A: Correlations
Variable CEO_INV CEO_INI
CEO_INI Pearson correlation Sig. (2-tailed)
1 0.947***
0.000 CEO_INV Pearson correlation Sig.
(2-tailed)
0.947*** 1
0.000 *** Correlation is significant at the 0.01 level (2-tailed)
N = 186 Panel B: Crosstabulation CEO_INV CEO_INI 0 1 Total 0 84 5 89 1 0 97 97 Total 84 102 186
Panel C: Chi-Square Tests
Value Df Asymptotic
Sig. (2-tailed)
Pearson Chi-square 166,945a 1 0.000
N of Valid Cases 186
27 is not involved in the fraud process. However, if an employee is involved than this employee is the initiator in 83 cases out of 117 cases. Therefore, hypothesis 1b is accepted.
Table 3 Panel A: Correlations
Variable EMP_INV EMP_INI
EMP_INV Pearson correlation Sig. (2 tailed)
1 0.689***
0.000 EMP_INI Pearson correlation Sig.
(2-tailed)
0.689*** 1
0.000 ***Correlation is significant at the 0.01 level (2-tailed)
N = 186 Panel B: Crosstabulation EMP_INV EMP_INI 0 1 Total 0 69 34 103 1 0 83 83 Total 69 117 186
Panel C: Chi-Square Tests
Value Df Asymptotic
Sig. (2-tailed)
Pearson Chi-square 88.393a 1 0.000
N of Valid Cases 186
28 Hypothesis 2. Table 4 shows the means of the number of techniques used in each phase of the fraud process. The mean of each phase demonstrates how many techniques are used in each phase on average. The number of techniques used increases in the initiation, proliferation and institutionalization phase when compared to the previous phase. This number declines in the detection phase compared to the institutionalization phase which can be explained by the number of observations. As explained above, the number of observations decreases in the proliferation and institutionalization phase since some fraud processes have no proliferation or institutionalization phase. The number of manipulation techniques used increases when more phases are included in the fraud process.
Table 4: Descriptive statistics
Variable Observations Mean Standard dev. Min Max
TECH_INI 186 1.457 0.721 0 5
TECH_PRO 172 1.814 0.937 0 5
TECH_INS 151 2.172 1.204 0 5
TECH_DET 186 2.000 1.286 0 5
To test hypothesis 2, T-tests are used to examine whether the manipulation method differs dependent on the phase in the fraud process. Paired sample t-tests are used to compare the differences in the number of manipulation techniques used between two phases of the fraud process13. The first paired sample t-test was conducted to compare the differences in the number of manipulation techniques used in the initiation and proliferation phase. The assumption of independence is met since the paired observations are drawn independently. The assumption of normality is not met since the Shapiro-Wilk is significant meaning that there is no normality, see table 5.
Table 5: Test of Normality
Kolmogorov-Smirnov Shapiro-Wilk
Variable Statistic Df Sig. Statistic Df Sig.
TECH_INI 0.391 151 0.000 0.649 151 0.000
TECH_PRO 0.240 151 0.000 0.855 151 0.000
TECH_INS 0.178 151 0.000 0.926 151 0.000
TECH_DET 0.171 151 0.000 0.927 151 0.000
Since the assumption of normality is violated, a Wilcoxon signed rank test can be used, instead of the paired sample t-test. This Wilcoxon signed rank test assumes no normality, independent observations and continuous variables. These assumptions are met. Table 6 shows the results
29 of this test. There was a significant difference in the number of techniques used in the initiation and proliferation phase, Z = -5.807, p < 0.001. There was a significant difference in the number of techniques used in the proliferation and institutionalization phase, Z = -2.085, p < 0.005 as well. There is no significant difference in the number of techniques used in the institutionalization and detection phase. Based on the described results, the number of manipulation techniques used differs dependent on the process stage. Therefore, this hypothesis can be accepted.
Table 6: Wilcoxon signed rank test
Z Sig (2-tailed) Pair 1: TECH_INI TECH_PRO -5.900 0.000*** Pair 2: TECH_PRO TECH_INS -3.828 0.000*** Pair 3: TECH_INS TECH_DET -0.496 0.620 *** Significant at the 0.01 level
Hypothesis 3. This hypothesis tests whether the manipulation method used in financial reporting fraud differs when other stakeholders contribute to this fraud process. There exists a significant correlation between the difference in manipulation methods used and the number of stakeholders contributing to the fraud process, r = 0.354, p < 0.001 (Table 7, Panel A). This hypothesis is tested by using logistic regressions. The assumption of independent observations is met. There is no hint for multicollinearity since all correlations are smaller than 0.8, see table 7. I therefore did not do any additional tests for multicollinearity. The assumption of sample size is met as well since the number of ten observations is significantly exceeded. Table 8 Panel A shows a significant association between the difference of stakeholders in the initiation and the proliferation phase and the difference between the manipulation techniques used in the initiation and proliferation phase, p < 0.001. Moreover, Panel B shows that there is a significant association between the difference of stakeholders in the initiation and the proliferation phase and the difference between the manipulation techniques used in the initiation and proliferation phase, p < 0.005. Panel C shows that the difference of stakeholders in the institutionalization and the detection phase has no significant association with the difference between the manipulation techniques used in the institutionalization and detection phase. I can conclude that the number of techniques used is influenced by an increased number of different stakeholders. This hypothesis can therefore be accepted.
30
Table 7: Correlations
Panel A: Between difference in techniques and stakeholders
DIFF_TECH DIFF_STAKE N
DIFF_TECH 1.000 0.354*** 186
DIFF_STAKE 0.354*** 1.000 186
Panel B: Proliferation phase
DIFF_TECH_PRO DIFF_STAKE_PRO N
DIFF_TECH_PRO 1.000 0.316*** 172
DIFF_STAKE_PRO 0.316*** 1.000 172
Panel C: Institutionalization phase
DIFF_TECH_INS DIFF_STAKE_INS N
DIFF_TECH_INS 1.000 0.278*** 151
DIFF_STAKE_INS 0.278*** 1.000 151
Panel D: Detection phase
DIFF_TECH_DET DIFF_STAKE_DET N
DIFF_TECH_DET 1.000 0.078 186
DIFF_STAKE_DET 0.078 1.000 186
*** Significant at the 0.01 level
Table 8: Logistic regressions Panel A: Proliferation phase
DIFF_TECH_PRO Odds Ratio Std. Err. z P>z 95% Conf. Interval DIFF_STAKE_PRO 2.110 0.463 3.40 0.001*** 1.372 3.244
_CONS 0.430 0.087 -4.19 0.000 0.290 0.638
Number of obs: 172 LR Chi2(1): 14.30 Prob>Chi2: 0.0002 Pseudo R2: 0.062 Log likelihood: -108.277
Panel B: Institutionalization phase
DIFF_TECH_INS Odds Ratio Std. Err. z P>z 95% Conf. Interval DIFF_STAKE_INS 1.864 0.414 2.80 0.005*** 1.206 2.881
_CONS 0.603 0.113 -2.70 0.007 0.418 0.871
Number of obs: 151 LR Chi2(1): 12.60 Prob>Chi2: 0.0004 Pseudo R2: 0.0601 Log likelihood: -97.408
Panel C: Detection phase
DIFF_TECH_DET Odds Ratio Std. Err. z P>z 95% Conf. Interval
DIFF_STAKE_DET 1.391 0.462 0.99 0.320 0.725 2.667
_CONS 0.008 0.007 -5.61 0.000 0.001 0.043
Number of obs: 186 LR Chi2(1): 0.72 Prob>Chi2: 0.396 Pseudo R2: 0.033 Log likelihood: -10.695
31 Hypothesis 4 and 5. A bar chart is used to provide a
visual overview of the number of stakeholders
involved in the fraud process (figure 1). Table 9 shows that the average number of stakeholders increases in each phase meaning that more stakeholders contribute to the fraud process when the duration of the fraud increases.
Table 9: Descriptive statistics
Variable Observations Mean Standard dev. Min Max
NUMB_INI 186 1.973 1.037 0 6
NUMB_PRO 172 2.442 1.394 0 10
NUMB_INS 151 2.722 1.919 0 10
NUMB_DET 186 2.866 1.761 0 10
I test these hypotheses by using paired sample t-tests to compare the number of stakeholders in the different phases of the fraud process. The assumption of independent observations is met since the paired observations are drawn randomly. The observation of normality is not met since the observations are not normally distributed (table 10).
Table 10: Test of Normality
Kolmogorov-Smirnov Shapiro-Wilk
Variable Statistic Df Sig. Statistic Df Sig.
NUMB_INI 0.245 151 0.000 0.816 151 0.000
NUMB_PRO 0.211 151 0.000 0.875 151 0.000
NUMB_INS 0.177 151 0.000 0.899 151 0.000
NUMB_DET 0.203 151 0.000 0.853 151 0.000
Since the assumption of normality is violated, a Wilcoxon signed rank test can be used to test whether there is a difference in the number of stakeholders involved. This test assumes no normality, independent observations and continuous variables. All of these assumptions are met. Table 11 demonstrates the results of Wilcoxon signed rank test. There exists a significant difference in the number of stakeholders involved used in the initiation and proliferation phase, Z = -5.807, p < 0.001. The results furthermore show that there is a significant difference in the number of stakeholders involved in the proliferation and institutionalization phase, Z = -2.085,
32 p < 0.005. There is a significant difference in the number of stakeholders involved in the institutionalization and detection phase, Z = -3.282, p < 0.001 as well. The number of stakeholders involved grows in the proliferation phase relative to the initiation phase meaning that hypothesis 4 can be accepted. Hypothesis 5 can be accepted as well since the number of stakeholders involved is almost stable in the institutionalization phase of the fraud process.
Table 11: Wilcoxon signed rank test
Z Sig (2-tailed) Pair 1: NUMB_INI NUMB_PRO -5.807 0.000*** Pair 2: NUMB _PRO NUMB _INS -2.085 0.037** Pair 3: NUMB _INS NUMB _DET -3.282 0.001*** ** Significant at the 0.05 level
*** Significant at the 0.01 level
Hypothesis 6. This hypothesis tests whether new stakeholders contribute to fraudulent financial reporting in the socialization phase. As explained in the methodology section, the institutionalization and socialization phases might be intertwined which makes it extremely difficult to distinguish these two phases in the complaints. I therefore combined these phases in one phase: the institutionalization phase. By checking the Q-Q and P-P plots it can be seen that the assumptions of linearity, normality and homoscedasticity are met. There is no multicollinearity since the VIF has a value of one. Table 12 shows that there are more stakeholders involved in the institutionalization phase. Stakeholders might contribute to fraudulent financial reporting in this phase by enlargement or replacement. I test this hypothesis by using a linear regression. Both contribution by replacement and contribution by enlargement have a significant effect on the involvement of new stakeholders in the institutionalization phase, p < 0.001. This leads to hypothesis 6 being accepted.
Table 12: Linear Regression NEW_SOC Coeff. Std. Err. T P>|t| [95%
conf.
interval] Tolerance VIF
NEW_REP NEW_ENL 0.773 0.612 0.054 0.051 14.39 12.00 0.000*** 0.000*** 0.667 0.511 0.880 0.713 1.000 1.000 1.000 1.000 _CONS 0.024 0.022 1.10 0.271 -0.019 0.068 1.000
*** Significant at the 0.01 level
33 Hypothesis 7. I observed whether an employee is the detector of financial reporting fraud in the detection phase. I observe this by looking at the means of the detector in each fraud case. Table 13 Panel A shows that most frauds (37.5% of the cases) are on average discovered by external auditors. Hypothesis 7 tests whether the number of stakeholders involved has a positive impact on internal detection. The R2 is extremely low (0.0072) which means that this model is not adequate for explaining the data. Panel B shows that there is no significant relationship between the number of stakeholders involved and internal detection. Therefore, this hypothesis cannot be accepted.
Table 13 Panel A: Descriptive Statistics
Variable Observations Mean Standard dev. Min Max
DET_INAUD 128 0.047 0.212 0 1 DET_EXAUD 128 0.375 0.486 0 1 DET_AUC 128 0.180 0.385 0 1 DET_EMP DET_MAN DET_VPS DET_MED DET_WHI DET_ANA DET_OTH DET_INT 128 128 128 128 128 128 128 128 0.250 0.203 0.008 0.000 0.031 0.000 0.164 0.273 0.435 0.404 0.088 0.000 0.175 0.000 0.372 0.447 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 Panel B: Regression
DET_INT Coeff. Std. Err. T P>|t| [95% conf. interval] NUMB_INV _CONS -0.009 0.299 0.026 0.081 -0.36 3.68 0.723 0.000 -0.060 0.138 0.042 0.459 N: 128 F(1,126): 0.13 Prob>F: 0.7228 R2: 0.0072 Adjusted R2: 0.703 MSE: 0.233
